Skip to content

Adding ciflow/h100-distributed when merging PR #6842

@huydhn

Description

@huydhn

Context

  • We have ciflow/h100-distributed label to run distributed tests on 8xH100 runner
  • Add auto labeling of H100 CI for distributed related changes #6833 was landed to add this label to all distributed PRs. However, the behavior of ciflow label is to run this job whenever the pull request is updated. This is too much and starts causing queue on the only 8xH100 runner we have

The ask

The ideal behavior describe by @kwen2501 and @fduwjj from distributed team is to add ciflow/h100-distributed only when merging the PR. This is similar to the land check concept we had for ciflow/trunk a while ago.

Is there a clear way to implement this logic? We could update the bot to do this, but it will need to incorporate the logic from #6833 to do so only for distributed pull requests instead of every PRs

    [/torch\/csrc\/distributed/gi, "ciflow/h100-distributed"],
    [/torch\/distributed/gi, "ciflow/h100-distributed"],
    [/torch\/testing\/_internal\/distributed/gi, "ciflow/h100-distributed"],
    [/test\/distributed/gi, "ciflow/h100-distributed"],

cc @clee2000 @malfet What are your thoughts on this?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    Status

    Cold Storage

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions